How to hack an AI

After going through a rather tumultuous exams season, and having taken a few weeks off to recharge batteries :slight_smile:, I finally seem to have found the time and the determination to get some real work done on Vena’s Tale (my Fractal Cosmos novel, for the ones who don’t know).

During the next month and a half, I won’t have much to do, so, I really hope I can make some actual progress with this.

Unfortunately, my writting has currently reached an impass and I frustratingly haven’t been able to add a single new paragraph to my story for quite sometime.

The reason for my impass is that storytelling is forcing me to tap into an area of which I frankly don’t know much about: computer science.

Knowing that the Fractal Future community certainly includes people who are well-versed in that field of science, I decided to turn to you guys for help :smiley: That is, if you don’t mind, of course. :slight_smile:

So, my problem is basically this:
My story has reached a point in which the protagonist (a 13 year-old posthuman girl named Vena) is currently standing at the top of tallest building in her city (which is not actually very tall, as her people has a preference for relatively short buildings because they look more retro) and she is being physically and verbally assaulted by a group of bullies from her school.
As she is heavily outnumbered and can’t fight, the only way she has to turn the odds in her favour is to hack into the building’s security and maintenance system (which is controlled by an advanced but subsentient near-AGI) in order to make it recognize her agressors as threats.

How would she do that? I want to provide the reader with some details to make it cool :smiley:, but I don’t want to be the sort of writer who resorts to technobabble just because he doesn’t totally grasp a certain field of science.

So, I need to write a relatively brief description of a cybernetically enhanced posthuman hacking an AI. Can you guys help me with that?

Any insights you could possibly give wpuld be very much appreciated. If you have any quick reading material that you think I should check out, I’d like that as well.

Thanks for you atention :slight_smile:


Good idea! :slight_smile:

My own knowledge of computer science is not very advanced, but perhaps I may offer some modest help.

This seems to boil down to creating a classifier that distinguishes between threats and non-threats. See:

As the system is probably advanced it would probably be a multi-step system that first classifies potential threats and then analyses them again to see whether they are actual threats. More likely there would be different classifiers for different kinds of threats.

There are at least two different ways of making an AI recognize the bullies as threat: Train it to identify those bullies as threat, or modify its sensory input to make the bullies appear as something else, for example fire or terrorists. The first approach would probably take a long time, because Vena would need to feed a training set of inputs that should, respectively should not be classified as threats. Such training sets need to be quite large to produce good results. Coming up with a large training set would probably take too much time, so the second approach would be much preferable.

Needless to say, messing with such an AI would be seen as a serious threat by itself, especially if the AI can initiate counter-measures on its own, instead of simply informing other people or systems about a potential problem. It’s also quite expected that a lot of students and penetration testers would want to mess with the AI of the building, so it’s probably not easy to find a weak spot in its security. So, it would be more plausible if Vena already tried to hack the AI systems of buildings before, if she’s to succeed with hacking it quickly.

There are two approaches that would lend themselves to classifying the bullies as threats:

  1. Hacking the sensors, so that they will present something that looks like a threat.
  2. Knowing how the bullies are classified by the AI exactly and adding a rule that entities like that are to be considered as threats.

In all likelihood the hack would be detected relatively quickly, but getting in trouble with the authorities may be preferable to be attacked by bullies. After all, the authorities might even be impressed by Vena’s hacking talents and could offer her a job!


Thanks for you input, Michael :slight_smile:

I’ll try to learn more about about the mechanics of statistical classification. That should be interesting.

It’ll definitly have to come down to modifying the system’s sensory input. The thing is that, for her to do that, she’ll problably have (I guess) to access the “code” of the AI, or maybe she just needs to tap into some sort of “control panel” (maybe she could crack the password or something…).

So, basically, I need to learn more about cryptography (as I understand it, these things generally come down to that…). I tried to find something to read on the topic, but I couldn’t find a good place to start.

Let’s see if anyone else has anything to say…

The AI’s ability to initiate “physical” countermeasures on its own is somewhat reduced because of the Venusians’ mild technophobia. It can lock down doors and stuff like that and it has control over a couple of security-drones that can immobilize people if necessary (but they are programmed not to harm anyone). It obviously can’t hack into people’s brains or anything like that (that would be ilegal, of course).

Sure, she does it all the time, she probably has a police record already :grinning:

ahahahah! Not quite what I have in mind, but very funny indeed.

It may actually be a good idea to have the police drop by. I may use this chance to introduce the detective character that I have in mind (he’ll act as a sympathetic antagonist in the first part of the story, persecuting Vena while she tries to build her giant robot, and I may actually think of making him a hero later on).

1 Like

in many science fiction stories you just need to know how to call an AI ( e.g. know its name or give it an authorization code). what about talking to it and convince it ( with logic) that the situation is of its concern and that it should interfere?

1 Like

Well…I’ve never really thought of this AI as the talking type, but I’ll keep that as option.

I’d still prefer old school hacking, thought.

But thanks for your input, you gave me a good plan B.


As Posthuman, surely she’d be able to interface with any AI using biotech grown organically as an adjunct to her neurophysiology.

Perhaps whale or dolphin DNA was modified to employ a sonar-like ability at a wave frequency far higher than soundwaves.

1 Like

She has brain implants. She certainly can interface with the AI just the same way a computer would do.

What I’m trying to figure out is what exactly she would do once she interfaced.

i did not understand the described situation. i did not think that in a post-human area there will be “bullies from her school” to assault post-humans.
it`s so strange like being assaulted by chimpanzes in a club-med.


Well, I guess we can say that early 22nd century Venus can indeed be a stange place at times. We cannot know for sure what the future will be like, but I don’t see any particular reason why schools or bullies would necessarily stop existing.

Anyway, I actually wrote this part of my story alredy. It’s the first chapter. If you’re interested, you can read it here: